You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
fix(gh-aw): finalize activation bindings and reconcile without creating issues - #2204
Add deterministic actual-ID activation finalization, explicit current-state no-create reconciliation, safe phase continuation, and SDK/checker consumption. Working as agentic-workflows-dev (Agentic Workflows Engineer); rejected-source revisions were independently authored. Merged #2202 remains unchanged.
Why
Closes#2203. Creating issues is insufficient if published bindings contain unresolved references or invalid identity. Duplicate creation is not recovery, and a concurrent run's valid comment cannot establish this run's publication.
How
Freeze and validate plan, registry, stable IDs, dependencies, authorization and complete outgoing issue-body bounds, including pinned sanitizer/attribution formatting. Retry-safe prerequisites precede admission; durable pending/start evidence precedes every create. Actual pinned runtime results drive finalization. Ambiguous creates never authorize recreation. Code-aware parsing prevents examples from becoming operational commands/tasks.
Preserve original records. Eligible hash-bound requests provide complete existing-issue mappings for current-state reconciliation, not historical attestation. Source/request/evidence phases must match. Expanding ancestry requires a nonempty, strictly shorter positional prefix; whole-plan predecessors cannot expand, and every explicitly selected phase must exist in the frozen plan. Sparse phase IDs remain valid. Validate prior maps, hashes, dependency closure and current API parents. Dependency-token corrections include readback; GitHub updates are not transactional.
The trusted deterministic writer validates actual publication receipts and API readbacks before reporting current-run completion. Historical collectors do not certify the completed run's publication. Automatic historical audits retain triggers but bound discovery to a 24-hour hint, five discovery pages, five origins and ten initial history pages per origin, with a shared 250-request ceiling across collection and canonical verification. Truncation/exhaustion/missing evidence fail explicitly. Manual full audit remains available; timestamps never establish current-run attribution.
Native installation packages 20 resources (19 shared plus the top-level checker) through supported staging and verified ownership/materialization; differing pre-existing checker files are preserved. No permission, credential, foreign-write, dispatch-policy or revision-trust-policy expansion.
Final reviewed head: f558617be67e51dc0ef8b6223e739e5369ec5b07; tree: 5536d8ed87bb726f937e949f3686b669caaec555. The 41-path scope is justified by the coupled producer/verifier/SDK/checker contract, native generated resources and regression coverage; no files deleted. Independent specialists corrected rejected source; the original author mechanically integrated, generated and validated it.
Final exact-head independent/hosted disposition: Security/recovery correctness APPROVE; Lead and Quality approve the narrow substitutions below. Hosted CI is green and all 11 inline threads are resolved. The final hosted keyword-casing usability observation is tracked separately in #2205: documented lowercase reconciliation works and unsupported spelling explicitly rejects. No remaining acceptance/recovery blocker; no gate waived for latency.
⚠️ Quick Check
If SDK/CLI source files changed: completed the applicable Changeset step below (npx changeset add / .changeset/*.md, direct CHANGELOG.md entry for maintainers, or skip-changelog label for no user-facing changes)
PR Readiness Checklist
The PR readiness bot will validate these automatically after push.
Check each item before requesting review. See CONTRIBUTING.md for full details.
Branch & Commit
Branch created from dev (not main)
Branch is up to date with dev (git fetch upstream && git rebase upstream/dev)
Verified diff contains only intended changes (git diff --cached --stat)
PR is not in draft mode (mark ready when checks pass)
Commit history is clean (squash fixups before review)
Live base is fc1cf596c1429f4c9ff7a99b90acd85d02cfdac2. Logical commits preserve independent attribution; no force push/history rewrite. Exact-head matching and live protections govern normal merge.
Build & Test
npm run build passes
npm test passes (all tests green)
npm run lint passes (type check clean)
npm run lint:eslint passes
For migration PRs (>20 files): include test output summary in PR description
npm test -- --run test/gh-aw-*.test.ts test/check-agent-binding.test.ts test/agent-provenance.test.ts test/recast-identity-neutrality.test.ts test/template-extra-classification.test.ts test/template-sync.test.ts
Passed: 36 suites / 3321 tests.
npm run build -w packages/squad-sdk
Independent Quality reviewer passed the documented prerequisite.
SQUAD_GH_AW_BIN=<absolute-gh-aw-v0.91.5-binary> npm test -- test/gh-aw-activation-finalization.test.ts test/agent-provenance.test.ts
Independent Quality reviewer passed 2 files / 459 tests after the SDK build.
npm test
Hosted final-head run passed 332 suites / 11,366 tests, 2 skipped and 47 todo.
Final CI run 38285719747 passed all nine jobs, including required gh-aw strict compile and test; isolation checks passed 443 tests before and after. Other returned repository checks succeeded; Scope Boundary intentionally skipped.
Compiler-dependent local commands use checksum-verified gh-aw v0.91.5 via PATH/SQUAD_GH_AW_BIN. Dependency manifests/lockfile unchanged. No additional standalone contract-export rerun is claimed for the final head: integrated fixtures already exercise actual native installation, strict compiled execution and the new end-to-end matrix.
The final matrix has 17 rows for activation and pending sources (34 actual-path cases), including genuine sparse [1] -> [1,3] -> [1,3,9] growth, ten verifier guard mutations, and no-create/map/hash assertions. Earlier retained regressions cover actual MCP serialization/sanitization and pinned create adapter, forged acceptance, authenticated legacy correspondence, malformed phases, complete formatted UTF-16 body boundaries, retry-safe prerequisites versus ambiguous creates, publication/readback, historical audit budgets and native ownership.
Final independent ancestry reviewer separately executed 30 end-to-end cases, ten isolated guard mutations and eight mapping/hash/topology groups. Verified built SDK emission, all 20 runtime-resource hashes and eight preserved compiled locks. That reviewer did not rerun installation/compilation/full CI; those are separately evidenced above. Actual hosted Copilot review 5494233472 completed on this exact head; COMMENTED alone is not treated as independent approval.
Changeset
Changeset added via npx changeset add (if packages/squad-sdk/src/ or packages/squad-cli/src/ changed)
Or direct CHANGELOG.md entry (maintainers only — write-protected for external contributors)
Or skip-changelog label applied (if no user-facing changes)
.changeset/activation-reconciliation.md uses the supported directly authored fragment alternative; generator not run. The unchecked entries are mutually exclusive alternatives, not missing changeset coverage.
Docs
README section updated (if new feature/module)
Docs feature page (if new user-facing capability)
Literal new README/feature-page additions are not claimed. Approved existing-guide substitutions are below. The guide provides public /casting usage, independent verifier responsibility, selected/undefined/error handling, recovery/phase/audit semantics, executable fixtures and SDK-build/pinned-tool instructions. Four new public interfaces have declaration JSDoc; category (c) is satisfied, not waived.
Exports
package.json subpath exports updated (if new module)
N/A: additions remain in the existing casting export; barrel updated, manifests unchanged.
Breaking Changes
Valid acceptance remains readable; malformed/unverified records reject. Activation requires stable IDs/frozen-plan fields rather than title inference or model-authored identity. Legacy promotion requires independent correspondence; missing evidence requires explicit current-state reconciliation. Later phases require validated prior evidence and ordered existing-phase growth. Reconciliation does not retroactively authenticate originals. Historical checker success does not certify run publication. The guide documents recovery and continuation; no existing public signature is removed.
Waived: category (d) separate SDK README feature section and new feature page under docs/src/content/docs/features/. Substitute the existing guide section Reading activation records with the SDK, reached through existing README/navigation links. No new page means conditional new-page navigation/docs-build-test additions do not trigger.
Waived: category (f) standalone sample project, separate sample README and sample-directory build/run form. Substitute actual executable fixtures test/gh-aw-activation-finalization.test.ts and test/agent-provenance.test.ts, the guide section Run the existing executable example, documented SDK build prerequisites and repository build/test evidence.
Reason: discoverable real producer/verifier/SDK regressions avoid a duplicate sample implementation. Approved at f558617be67e51dc0ef8b6223e739e5369ec5b07 by architect (Lead reviewer 608a9e2f-a904-495a-8854-00a97d1dcf10) and integration-e2e-tester (Quality reviewer c6abbe01-9613-483a-a85c-d5ac72c06de2), recorded by the coordinator from independent reports. Quality independently ran the documented build plus 459-test example. These are location/harness substitutions, not waived coverage. Public API JSDoc, runtime regressions, independent source/security review, hosted CI, protections and merge gates remain unwaived.
Template files changed in .github/workflows/ but not in other template locations. If these templates should stay in sync, consider updating the others too.
The reason will be displayed to describe this comment to others. Learn more.
🟡 Changes recommended
Pre-create failures can permanently strand activation, oversized accepted plans can exceed GitHub’s body limit, and the SDK accepts malformed legacy phase scopes.
Revision independently authored by github-integration-dev under reviewer lockout. Address legacy promotion authority, full-body acceptance forgery, phase-scope parity, rendered body bounds, and retry-safe prerequisites. Original author performed only validation and mechanical integration.
Closes#2203
Revision-Author: github-integration-dev
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 9370fe46-c2d1-486c-ab53-dc21c7f4d0fb
Stage native YAML through the supported .github/aw package destination and materialize its verified workflow without overwriting unrelated content. Exercise the actual installer and preserve canonical label/isolation gates. Keep topology at nineteen shared resources plus one native checker.
Closes#2203
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 9370fe46-c2d1-486c-ab53-dc21c7f4d0fb
In chain mode this compares consumed activation/source records with every caller-supplied record. A normal discussion comment is ignored in single-record mode but makes an otherwise valid multi-phase chain fail as “unconsumed history.” Count only classified activation/pending history records here, while continuing to reject unconsumed activation candidates.
Compare chain consumption against classified history records
workflows/shared/check-agent-binding.mjs:919
This chain-consumption check counts unrelated input comments even though the selector's classification intentionally skips them. Consequently adding a discussion-only record breaks a valid multi-phase selection, unlike the single-pair path. Compare against the classified activation/pending history set rather than raw records.length.
Reconcile approved path count with the actual 41-file diff
workflows/squad-workflows.manifest.json:139
The PR description says this ships “exactly the approved 39 paths,” but the current PR changes 41 files. Please reconcile the stated scope/count with the actual diff so reviewers can verify that no unapproved paths were added.
Collector feasibility update for unchanged head cfb2e9649e3858c1acd88917bed9e12895edc97b:
The independent architect completed the authorized read-only deployed-capability assessment. All eight probes of the documented step-specific log endpoint returned HTTP 404, including known successful pinned upload steps and adjacent/index-offset candidates. Same-access exact-attempt job metadata, artifact metadata and a whole-job log returned HTTP 200. This establishes that the step-specific route was unusable with the tested access; it does not distinguish unsupported deployment from endpoint-specific access or data behavior.
The public artifact API does not expose the uploader job/attempt join. Matching artifact ID/digest text in whole-job logs is not a replacement for server-attributed trusted-step evidence. Existing candidate run 38198933531 is Squad CI, has no relevant receipt artifacts, and is not a native activation/reconciliation execution.
A separate source-authority decision remains: native workflow_dispatch currently permits an authorized branch/tag selection, so a default-only or revision allowlist would be an explicit behavior/policy change, not a silent repair. No such change has been made.
Both collector findings remain unresolved. No workflow/source/permission changes, new fixtures, dispatches or additional reruns were performed by this assessment. The passing same-head CI retry remains recorded separately and is not security or collector clearance. This PR remains draft and blocked pending a supported provenance architecture and authorized implementation followed by independent exact-head review.
Independent documentation/sample dispositions for exact head 4438fe716b290842f04b7a8d78bf9e44a623ad84, recorded by the source coordinator from reviewer reports:
Under .github/PR_REQUIREMENTS.md, both roles approve only these explicit substitutions:
Category (d): the existing gh-aw guide section Reading activation records with the SDK, reached through existing README/navigation links, replaces a separate SDK README feature section and a new feature page under docs/src/content/docs/features/. No new page is added, so conditional new-page navigation/docs-build-test additions do not trigger.
Category (f): actual executable fixtures in test/gh-aw-activation-finalization.test.ts and test/agent-provenance.test.ts, the guide instructions Run the existing executable example, and repository build/test evidence replace a standalone sample project, separate sample README, and sample-directory build/run form.
These are location/harness substitutions, not omitted coverage. The guide names the public /casting API, independent verifier-evidence boundary, selected/undefined/error handling, real producer/verifier/SDK fixture, pinned gh-aw v0.91.5 prerequisites and runner. Quality independently executed the documented SQUAD_GH_AW_BIN=<absolute-gh-aw-v0.91.5-binary> npm test -- test/gh-aw-activation-finalization.test.ts test/agent-provenance.test.ts: 364 tests passed across 2 files. Both reviewers confirmed the four new public interfaces now have declaration JSDoc; category (c) is satisfied, not waived.
The earlier documentation/sample rejections at 5142706f are cleared for this exact scope. This is not overall PR approval, security clearance or permission to bypass required checks. Runtime regressions, independent source/generated review, changeset, hosted CI, branch protections and normal merge gates remain unwaived.
Update troubleshooting guidance to prevent duplicate activation recovery
docs/src/content/docs/guide/gh-aw.md:1866
This no-retry rule conflicts with the existing troubleshooting guidance at lines 852 and 2570-2577, which still tells operators to rerun activation, resume by title, and rely on the old 75/110 caps and report_incomplete behavior. Following those instructions after an ambiguous create can cause the duplicate recovery this section prohibits; update those sections to direct users to explicit current-state reconciliation and the current limits.
Use the existing canonical pinning helper before compiling; preserve the exact immutable SHA assertion. Separate mechanical fixture correction for Closes#2203.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 9370fe46-c2d1-486c-ab53-dc21c7f4d0fb
Document the SDK build prerequisite for both supported compiler paths. Independently revised and exercised by docs-devrel against freshly generated public casting exports. Closes#2203.
Revision-Author: docs-devrel (42cb5949-3981-4e7a-a012-70862d711675)
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 9370fe46-c2d1-486c-ab53-dc21c7f4d0fb
Independent documentation/sample dispositions for exact head d97b1e6ae73ee7c3608f96660c232df5fb7c8b1b, recorded by the source coordinator from renewed reviewer reports:
Under .github/PR_REQUIREMENTS.md, both roles approve only these explicit substitutions:
Category (d): the existing gh-aw guide section Reading activation records with the SDK, reached through existing README/navigation links, replaces a separate SDK README feature section and a new feature page under docs/src/content/docs/features/. No new page is added, so conditional new-page navigation/docs-build-test additions do not trigger.
Category (f): actual executable fixtures in test/gh-aw-activation-finalization.test.ts and test/agent-provenance.test.ts, the guide instructions Run the existing executable example, and repository build/test evidence replace a standalone sample project, separate sample README, and sample-directory build/run form.
These are location/harness substitutions, not omitted coverage. The guide names the public /casting API, independent verifier-evidence boundary, selected/undefined/error handling, real producer/verifier/SDK fixture, pinned gh-aw v0.91.5 prerequisites and runner. The final correction explicitly requires npm run build -w packages/squad-sdk before BOTH normal and alternate-binary runners. Quality independently executed that build, then SQUAD_GH_AW_BIN=<absolute-gh-aw-v0.91.5-binary> npm test -- test/gh-aw-activation-finalization.test.ts test/agent-provenance.test.ts: build passed; 407 tests passed across 2 files. Four new public interfaces retain declaration JSDoc; category (c) is satisfied, not waived.
The earlier discoverability and missing-build-prerequisite findings are cleared for this exact scope and head. This is not overall PR approval, security clearance or permission to bypass required checks. Runtime regressions, independent source/generated review, changeset, hosted CI, branch protections and normal merge gates remain unwaived.
Reviewer: security-review agent 452066a4-25d5-4a5d-9d5b-82d9d81ed6f1.
Final immutable candidate: d97b1e6ae73ee7c3608f96660c232df5fb7c8b1b.
Verdict: APPROVE. Coordinator-preserved receipt from the independent reviewer.
The reviewer approved runtime candidate da46d7ccaea9440600e6377bd7bf4406512ff486 after 43 focused independent probes, then independently verified the complete guide-only delta to d97b1e6. No runtime, generated, package or security-policy changes occurred in that final delta. Probes were not redundantly rerun for the documentation-only commit.
Cleared findings and regression evidence
Seven code-example styles underwent two pinned sanitizer passes and source plus emitted compiled-job execution. Examples contributed no phases/tasks; legitimate tasks were created and finalized.
Seven source/request phase combinations ran through source/compiled reconciliation. Whole-plan omitted/empty scopes succeed; neither widens a scoped source; exact matches succeed; mismatches reject without mutation.
SDK and canonical checker accept ordinary discussion mixed with valid pair/chain history while rejecting malformed authority, extra activation candidates, malformed pending records, duplicate identities and missing proof.
Actual pinned compiled issue creation accepts complete outgoing bodies of 65,535 and 65,536 UTF-16 units; 65,537 rejects before any mutation for ASCII and supplementary Unicode. Missing formatter rejects before mutation.
Critical runtime bytes were independently compared to immutable public pin 16b430146d5d5646eccef5a1ceb72152333b18b9.
Limits
The reviewer reused previously reviewed baseline evidence and focused on revision/regression deltas. It did not independently rerun the full suite, fresh strict compilation, actual native installer/materialization, or hosted end-to-end execution. Those execution gates remain separately attributed to integration/hosted evidence. No static suite totals were substituted for this security verdict.
No GitHub or repository mutations were performed by the reviewer. Coordinator owns final exact-head CI, hosted review, documentation/sample disposition and normal merge qualification.
This now performs an unbounded repository-wide comment scan after every Squad/Squad CI completion. The checker then makes additional API reads for each historical record and each referenced issue, so repositories with a few hundred activations can exhaust the 5,000-request token budget and make every subsequent audit fail. Keep full-history audits scheduled/manual, or pass a bounded candidate-origin set from the triggering run and API-verify the complete chain only for those origins.
Consumer runs scan all history and exceed the API budget
The installed checker has the same unbounded scan: every consumer Squad run paginates all repository comments and then re-reads issues and ancestry for every historical activation. This eventually exceeds the workflow token's API budget and turns unrelated Squad runs red. Scope per-run audits to bounded candidate origins (while API-verifying their full chains), and move repository-wide history audits to a scheduled/manual workflow.
Require source activation phase scope to match verified evidence in both SDK and canonical checker. Add built-SDK parity and defensive mismatch regressions without changing public APIs or trusted writer behavior. Closes#2203.
Revision-Author: typescript-api-contract-designer (c2726384-189e-4dfe-a10f-f466007b4852)
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Copilot-Session: 9370fe46-c2d1-486c-ab53-dc21c7f4d0fb
Candidate: c2b3cc9867f57f2dac0a91653cbd8005e11d4aa5.
These decisions are recorded by the source coordinator from independent reviewer reports; they do not replace hosted checks or branch protection.
Security and contract review: APPROVE, reviewer 452066a4-25d5-4a5d-9d5b-82d9d81ed6f1.
The final paired consumer-scope correction passed ten focused scenario groups using the actual built SDK public export, exact-source SDK and canonical selector. Source [1] rejects replacement/evidence [1,2], [] and [2], including fabricated evidence. Matched phase, whole-plan explicit/omitted evidence, pending recovery, complete two-phase chains and ordinary discussion remain accepted. Extra/malformed activation authority rejects. The reviewer classified the former consumer gap as LOW defensive consistency: the actual trusted producer/verifier already rejected widening, so HIGH authorization impact was not demonstrated.
The reviewer independently inspected the final six-file delta and bound built SDK JavaScript to exact source (excluding only source-map directive). It verified 20 source/installed runtime-resource digests, exported manifest, eight workflow-source hashes and eight normalized compiled locks with immutable pin checks. Writer authority, permissions, ownership and revision policy are unchanged.
Prior independent revision evidence remains valid for unchanged code: 43 focused probes cleared parser, scope, ordinary-history, complete outgoing-body attribution/UTF-16 limits and ambiguity protection; 19 probes cleared bounded historical audit behavior, including the shared 250-request ceiling across discovery and canonical verification, pagination/truncation failures, both actual YAML scripts and manual full-audit behavior. Discovery hints never establish current-run attribution.
Limits: these reviews used synthetic API/virtual-filesystem probes and hash-qualified built/exported artifacts. The reviewer did not independently execute a fresh build, installer/materializer, compiler, hosted run or full suite. Integration and hosted evidence cover those separate gates; suite totals were not substituted for review.
Lead documentation/sample substitutions: APPROVE, architect reviewer 608a9e2f-a904-495a-8854-00a97d1dcf10.
Both approvals apply to the exact candidate and only these substitutions under .github/PR_REQUIREMENTS.md:
Category (d): existing gh-aw guide/public casting example and README/navigation links replace a separate SDK README feature section and new feature page. Conditional new-page navigation/tests do not trigger.
Category (f): named production-path executable fixtures, guide prerequisites/build/test/results handling and repository validation replace a standalone sample project, separate sample README and sample-directory build/run form.
Quality independently ran npm run build -w packages/squad-sdk, then SQUAD_GH_AW_BIN=<absolute-gh-aw-v0.91.5-binary> npm test -- test/gh-aw-activation-finalization.test.ts test/agent-provenance.test.ts: build passed; 423 tests passed across two files. Public API/JSDoc, signatures, current-build prerequisite and guide contract remain intact. Category (c) is satisfied, not waived.
No runtime coverage, independent review, full CI, changeset, permissions, protection or merge gate is waived.
The chain accepts non-prefix ancestry because it only checks set inclusion. For example, independently evidenced scopes [2] → [1, 2] pass here, although the producer verifier requires prior phases to occupy the same leading positions (squad-activation.mjs:1376-1378). This lets the public selector recognize a phase history that the trusted workflow explicitly rejects; compare phases by index so ancestry is an exact expanding prefix.
Enforce exact prefix ancestry in canonical checker
workflows/shared/check-agent-binding.mjs:968
This canonical checker has the same set-inclusion gap as the SDK selector: [2] → [1, 2] is treated as expanding ancestry even though the trusted request verifier requires the predecessor scope to be an exact prefix (squad-activation.mjs:1376-1378). Enforce positional prefix equality here so checker and producer contracts cannot diverge.
The reason will be displayed to describe this comment to others. Learn more.
🔵 Needs a closer look
The large security-sensitive workflow change still has a command parsing defect and awaits exact-head CI and security qualification.
0 open findings
Previously missed (1)
In code that hasn't changed since last review
Match squad command words case-insensitively
workflows/shared/squad-command-contract.mjs:182
This new command is the only argument classifier that matches normalized case-sensitively; all existing commands classify through lowered. As a result, /squad Activate is accepted but /squad Reconcile Activation … is rejected. Match the command words case-insensitively while retaining the lowercase-only SHA requirement.
Independent reviewer activation-final-security APPROVES the consolidated ancestry delta from c2b3cc9. All three conditions behind the previous request for changes are cleared: non-prefix insertion, whole-plan predecessor expansion, and nonexistent selected phases.
#
Severity
File
Lines
Vulnerability
Confidence
-
-
-
-
No outstanding vulnerability found in the scoped final review
-
The reviewer independently executed 30 end-to-end activation/pending cases, 10 isolated guard-mutation controls, and 8 mapping/hash/topology regression groups through the actual built SDK public export, exact-source producer/verifier, and canonical package-source selector/CLI. Valid ordered and sparse growth, standalone whole-plan recovery, ordinary discussion, dependency closure, and no-create reconciliation remain supported. Changed mappings, broken ancestry, and competing histories reject.
The reviewer inspected the entire seven-file delta, compared built SDK JavaScript with exact-source emission, and checked 20 runtime-resource hashes, workflow/skill hashes, unchanged ownership/revision policy, and eight preserved compiled locks against the final manifest/runtime pins.
Limits: no fresh installer, compiler, hosted execution, or full-suite rerun by this reviewer; probes used virtual APIs/filesystem mutations. This explicit final-delta approval carries forward the previously recorded independent source/security and bounded-audit dispositions. Hosted totals were not used as substitute clearance.
Lead and Quality documentation/sample approvals
Both independent role reviewers APPROVE at the exact head:
architect, Lead (608a9e2f-a904-495a-8854-00a97d1dcf10): unchanged public signatures/JSDoc, consumer guide, fresh-SDK-build prerequisite, ordered/full/scoped guidance, changeset, and linked ancestry matrix remain adequate.
integration-e2e-tester, Quality (c6abbe01-9613-483a-a85c-d5ac72c06de2): independently inspected the delta and documented journey, then ran npm run build -w packages/squad-sdk followed by SQUAD_GH_AW_BIN=<absolute-gh-aw-v0.91.5-binary> npm test -- test/gh-aw-activation-finalization.test.ts test/agent-provenance.test.ts: build passed; 2 files / 459 tests passed.
Approved substitutions under .github/PR_REQUIREMENTS.md:
(d) Existing guide/public /casting usage with existing README/navigation links instead of a separate README feature section and new feature page.
(f) Named executable repository fixtures plus documented prerequisites, SDK build, test commands and result handling instead of a standalone sample project, sample README and sample-directory build/run.
Category (c) public API JSDoc is satisfied, not waived. Coverage, security, CI, runtime behavior and merge protections are not waived. Reports are recorded here by the coordinator, not represented as GitHub-native reviewer approvals.
Hosted review and CI
Exact-head CI run 38285719747 succeeded: all nine jobs; 332 suites / 11,366 tests passed, 2 skipped, 47 todo; isolation 443 tests before and after.
Actual hosted Copilot review 5494233472 completed on the exact head. All 11 inline review threads are resolved.
The final overview's command-keyword casing observation is deferred to Normalize reconciliation command keyword casing without relaxing SHA validation #2205. The actual classifier accepts documented lowercase recovery, explicitly rejects mixed-case reconciliation keywords and uppercase hashes, and does not authorize a partial operation. Independent security review agrees it is not a security blocker. This is a bounded usability follow-up, not a waived acceptance/recovery defect.
Landing disposition
No outstanding acceptance/recovery blocker. Final live read confirms head above, base fc1cf596c1429f4c9ff7a99b90acd85d02cfdac2, 41 paths with no deletions, CLEAN merge state, all returned checks successful or intentionally skipped, required up-to-date compile/test gates satisfied, and zero required GitHub approval count. Normal merge only, with exact-head matching; no bypass, auto-merge, permission expansion, deployment, or consumer adoption.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
Add deterministic actual-ID activation finalization, explicit current-state no-create reconciliation, safe phase continuation, and SDK/checker consumption. Working as agentic-workflows-dev (Agentic Workflows Engineer); rejected-source revisions were independently authored. Merged #2202 remains unchanged.
Why
Closes #2203. Creating issues is insufficient if published bindings contain unresolved references or invalid identity. Duplicate creation is not recovery, and a concurrent run's valid comment cannot establish this run's publication.
How
Freeze and validate plan, registry, stable IDs, dependencies, authorization and complete outgoing issue-body bounds, including pinned sanitizer/attribution formatting. Retry-safe prerequisites precede admission; durable pending/start evidence precedes every create. Actual pinned runtime results drive finalization. Ambiguous creates never authorize recreation. Code-aware parsing prevents examples from becoming operational commands/tasks.
Preserve original records. Eligible hash-bound requests provide complete existing-issue mappings for current-state reconciliation, not historical attestation. Source/request/evidence phases must match. Expanding ancestry requires a nonempty, strictly shorter positional prefix; whole-plan predecessors cannot expand, and every explicitly selected phase must exist in the frozen plan. Sparse phase IDs remain valid. Validate prior maps, hashes, dependency closure and current API parents. Dependency-token corrections include readback; GitHub updates are not transactional.
The trusted deterministic writer validates actual publication receipts and API readbacks before reporting current-run completion. Historical collectors do not certify the completed run's publication. Automatic historical audits retain triggers but bound discovery to a 24-hour hint, five discovery pages, five origins and ten initial history pages per origin, with a shared 250-request ceiling across collection and canonical verification. Truncation/exhaustion/missing evidence fail explicitly. Manual full audit remains available; timestamps never establish current-run attribution.
Native installation packages 20 resources (19 shared plus the top-level checker) through supported staging and verified ownership/materialization; differing pre-existing checker files are preserved. No permission, credential, foreign-write, dispatch-policy or revision-trust-policy expansion.
Final reviewed head:
f558617be67e51dc0ef8b6223e739e5369ec5b07; tree:5536d8ed87bb726f937e949f3686b669caaec555. The 41-path scope is justified by the coupled producer/verifier/SDK/checker contract, native generated resources and regression coverage; no files deleted. Independent specialists corrected rejected source; the original author mechanically integrated, generated and validated it.Final exact-head independent/hosted disposition: Security/recovery correctness APPROVE; Lead and Quality approve the narrow substitutions below. Hosted CI is green and all 11 inline threads are resolved. The final hosted keyword-casing usability observation is tracked separately in #2205: documented lowercase reconciliation works and unsupported spelling explicitly rejects. No remaining acceptance/recovery blocker; no gate waived for latency.
npx changeset add/.changeset/*.md, directCHANGELOG.mdentry for maintainers, orskip-changeloglabel for no user-facing changes)PR Readiness Checklist
Branch & Commit
dev(notmain)dev(git fetch upstream && git rebase upstream/dev)git diff --cached --stat)Live base is
fc1cf596c1429f4c9ff7a99b90acd85d02cfdac2. Logical commits preserve independent attribution; no force push/history rewrite. Exact-head matching and live protections govern normal merge.Build & Test
npm run buildpassesnpm testpasses (all tests green)npm run lintpasses (type check clean)npm run lint:eslintpassesSKIP_BUILD_BUMP=1 npm run buildnpm run lintnpm run lint:eslintnpm run gh-aw:integrity:writenpm run gh-aw:integrity:checknpm test -- --run test/gh-aw-*.test.ts test/check-agent-binding.test.ts test/agent-provenance.test.ts test/recast-identity-neutrality.test.ts test/template-extra-classification.test.ts test/template-sync.test.tsnpm run build -w packages/squad-sdkSQUAD_GH_AW_BIN=<absolute-gh-aw-v0.91.5-binary> npm test -- test/gh-aw-activation-finalization.test.ts test/agent-provenance.test.tsnpm testFinal CI run 38285719747 passed all nine jobs, including required gh-aw strict compile and test; isolation checks passed 443 tests before and after. Other returned repository checks succeeded; Scope Boundary intentionally skipped.
Compiler-dependent local commands use checksum-verified gh-aw v0.91.5 via
PATH/SQUAD_GH_AW_BIN. Dependency manifests/lockfile unchanged. No additional standalone contract-export rerun is claimed for the final head: integrated fixtures already exercise actual native installation, strict compiled execution and the new end-to-end matrix.The final matrix has 17 rows for activation and pending sources (34 actual-path cases), including genuine sparse
[1] -> [1,3] -> [1,3,9]growth, ten verifier guard mutations, and no-create/map/hash assertions. Earlier retained regressions cover actual MCP serialization/sanitization and pinned create adapter, forged acceptance, authenticated legacy correspondence, malformed phases, complete formatted UTF-16 body boundaries, retry-safe prerequisites versus ambiguous creates, publication/readback, historical audit budgets and native ownership.Final independent ancestry reviewer separately executed 30 end-to-end cases, ten isolated guard mutations and eight mapping/hash/topology groups. Verified built SDK emission, all 20 runtime-resource hashes and eight preserved compiled locks. That reviewer did not rerun installation/compilation/full CI; those are separately evidenced above. Actual hosted Copilot review 5494233472 completed on this exact head;
COMMENTEDalone is not treated as independent approval.Changeset
npx changeset add(ifpackages/squad-sdk/src/orpackages/squad-cli/src/changed)CHANGELOG.mdentry (maintainers only — write-protected for external contributors)skip-changeloglabel applied (if no user-facing changes).changeset/activation-reconciliation.mduses the supported directly authored fragment alternative; generator not run. The unchecked entries are mutually exclusive alternatives, not missing changeset coverage.Docs
Literal new README/feature-page additions are not claimed. Approved existing-guide substitutions are below. The guide provides public
/castingusage, independent verifier responsibility, selected/undefined/error handling, recovery/phase/audit semantics, executable fixtures and SDK-build/pinned-tool instructions. Four new public interfaces have declaration JSDoc; category (c) is satisfied, not waived.Exports
N/A: additions remain in the existing casting export; barrel updated, manifests unchanged.
Breaking Changes
Valid acceptance remains readable; malformed/unverified records reject. Activation requires stable IDs/frozen-plan fields rather than title inference or model-authored identity. Legacy promotion requires independent correspondence; missing evidence requires explicit current-state reconciliation. Later phases require validated prior evidence and ordered existing-phase growth. Reconciliation does not retroactively authenticate originals. Historical checker success does not certify run publication. The guide documents recovery and continuation; no existing public signature is removed.
Waivers
Final exact-head Lead and Quality approval record, under
.github/PR_REQUIREMENTS.md:docs/src/content/docs/features/. Substitute the existing guide section Reading activation records with the SDK, reached through existing README/navigation links. No new page means conditional new-page navigation/docs-build-test additions do not trigger.test/gh-aw-activation-finalization.test.tsandtest/agent-provenance.test.ts, the guide section Run the existing executable example, documented SDK build prerequisites and repository build/test evidence.Reason: discoverable real producer/verifier/SDK regressions avoid a duplicate sample implementation. Approved at
f558617be67e51dc0ef8b6223e739e5369ec5b07by architect (Lead reviewer608a9e2f-a904-495a-8854-00a97d1dcf10) and integration-e2e-tester (Quality reviewerc6abbe01-9613-483a-a85c-d5ac72c06de2), recorded by the coordinator from independent reports. Quality independently ran the documented build plus 459-test example. These are location/harness substitutions, not waived coverage. Public API JSDoc, runtime regressions, independent source/security review, hosted CI, protections and merge gates remain unwaived.